Normalize Retrieval Embeddings

Problem

Normalize each retrieval embedding to unit L2 length. Preserve zero vectors as zero vectors. Return a matrix with the original shape.

Compute norms per row, not across the whole collection; no epsilon is added to nonzero norms.

def normalize_embeddings(embeddings):

Constraints

• embeddings is a nonempty N × D matrix with 1 ≤ N,D ≤ 32.
• Finite entries have magnitude at most 1000.
• Inputs satisfy these constraints; invalid-input handling is not graded. Use only Python standard library and NumPy.

Examples

Arguments: [
  [
    [
      3,
      4
    ]
  ]
]
Expected result: [
  [
    0.6,
    0.8
  ]
]

Arguments: [
  [
    [
      0,
      0
    ],
    [
      0,
      -2
    ]
  ]
]
Expected result: [
  [
    0,
    0
  ],
  [
    0,
    -1
  ]
]